The Agentic Enterprise. AK · Daily Edition · 8 min read | Monday, November 2, 2026 Anthropic changes its IPO filingAnthropic's amended IPO filing shows $17 billion in third-quarter revenue, growth that slowed to 47% and an operating loss the company blames on pricier short-term compute. The first S-1 stopped at June. Monday's amendment fills in the quarter investors were waiting for, and it is less tidy than the one before it. Revenue still rose by $5.4 billion in three months. Expenses rose faster, by $6.9 billion, on pricier short-term compute and a $2 billion non-cash charge. If your company runs on Claude, this is the most detailed look at a model vendor's economics you have had, and it reads best as a supplier review. | The Big Story Vendors Anthropic's growth met its compute bill| A | nthropic's amended S-1, posted to the SEC on Monday and reported by Fortune, adds the third quarter the original filing left blank. Revenue reached $17 billion for the three months to September 30, up from $11.6 billion in the second quarter and $1.4 billion a year earlier. Sequential growth slowed to about 47%, from roughly 144% the quarter before. Operating expenses climbed to $17.5 billion, and a $974 million operating profit in the second quarter turned into a $567 million loss. Fortune · CNBC |
Anthropic attributes the swing mainly to compute bought under short-term commitments, which it says "currently costs significantly more" than capacity under its multi-year deals. A $2 billion non-cash charge for an employee donation-matching program is larger than the loss on its own. On an adjusted basis the company stayed profitable, though its adjusted operating margin fell to 15% from 19%. The net loss was $23.3 billion, mostly an accounting charge tied to the rising value of its convertible notes. Behind all of it sits about $895.4 billion in committed spending, mostly compute and infrastructure, against $146.6 billion in liquidity. Fortune · Original S-1 For buyers, one line matters more than the loss. Price cuts on new model versions and contractual discounts took about $600 million off revenue between the second and third quarters. The discount half of that tells buyers the price is negotiable. The original filing, which left the third-quarter revenue figure blank, already said models released in the prior six months made up more than 80% of it. The model you tested in the spring is probably not the one you are paying for now. Fortune · TheNextWeb A vendor paying a premium for short-term compute is a vendor whose costs can turn up in your next renewal. The Spearhead Take Read the prospectus the way procurement would. Ask whether your contract prices a unit of work or only a token, and who carries the cost when your vendor runs short of capacity. Anthropic is a Spearhead partner; we would give the same advice about any lab that filed these numbers. |
Sources: Fortune · CNBC · Original S-1 · TheNextWeb | The Obvious & The OverlookedWhat is easy to see, and what is easy to miss. The Obvious Compute is the gating input. The biggest clouds are racing to secure capacity and chips for agent workloads. Source: NetworkWorld. |
Spending keeps climbing. Gartner now sees worldwide AI spending near $2.67 trillion this year. Source: Techstrong.ai. |
|
The Overlooked Your cloud commit may already be your AI contract. Cloud marketplaces carried 48% of Anthropic's first-half revenue, up from 11% in 2023. Source: TheNextWeb. |
The cost center moved from training to running. For the first time, enterprises will spend more on inference than on training models. Source: Gartner. |
Even the hyperscalers hedge their chip supplier while buyers concentrate. Oracle will sell Google's TPUs; AWS and Google are diversifying away from a single accelerator. Source: The Stack. |
The adoption numbers measure copilots, not agents. Studies that report a majority in production rarely agree on what an agent is. Source: CMSWire on Liferay. |
|
|
| Moving PiecesFive developments with practical consequences. Infrastructure Oracle will sell Google's chips in its own cloudOracle said it will offer Google's TPUs on Oracle Cloud Infrastructure starting in 2027, the first time Google's custom AI chips will run in a hyperscale cloud other than Google's own. Oracle handles the power, cooling and networking; the chips come in training and inference variants. The point for buyers is optionality: a second serious accelerator to price against Nvidia, inside a cloud many enterprises already run. The catch is timing and a rival. This is future capacity, not something you can provision now, and Google is also building a TPU cloud with Blackstone, so Oracle may not be first in practice. NetworkWorld · The Stack | Deals A vertical agent company raised at $4 billionEliseAI, which builds AI agents for housing and healthcare operators, raised a reported $350 million Series F at a $4 billion valuation. The shape matters more than the number. Money is flowing to companies that own a narrow operational workflow end to end rather than to another general assistant. For a CIO, vertical agents are the easier buy: they arrive knowing the domain, and the integration surface is smaller. Whether the moat holds once the platform vendors ship the same workflow is the open question. New Economies | Markets Apple reports today, the megacap without the capex storyApple releases fiscal fourth-quarter results after the close, with analysts expecting roughly $113 billion in revenue and iPhone around $56 billion. It is the one megacap whose quarter is not dominated by an AI capital-spending story. While Microsoft, Alphabet, Amazon and Meta spend past their cash flow on data centers, Apple has largely sat out the buildout and leaned on partners for frontier models. For enterprise buyers, the quarter is a live test of whether restraint on AI capex reads as discipline or as falling behind. AppleInsider · TipRanks | Workforce The "AI-first" layoff keeps arriving, and so does the walk-backDropbox said it would cut about 9% of staff as part of an AI-driven restructuring, one of several enterprise software firms trimming headcount this fall and crediting efficiency. Set against it is a Gartner projection that by 2029 roughly 30% of employees replaced by AI will be rehired, often at higher cost, because the automation did not cover the job. Watch how many of this fall's cuts come back over the next two years. A rehire at a premium is the receipt for cutting staff to a tool that could not yet do the work. InformationWeek · ITPro | Deployment An oil major signed up for agents across its value chainADNOC signed a multi-year deal with OpenAI and G42 to run coordinated agents across oilfield planning, production, refining and trading, and will be the first energy company on OpenAI Frontier, the vendor's platform for linking agents across a business. The agents run on G42 infrastructure inside the UAE, which addresses the data-sovereignty question, at least on paper, and they land on a ChatGPT rollout that already reached 40,000 employees. People keep the consequential calls. Other operators should copy the order of work: seats and usable data first, agents second. Reuters via Yahoo Finance · Energy Connects |
| On the RadarEight signals worth keeping in view. Deals OpenAI signed four Abu Dhabi agreements at once. Beyond ADNOC, OpenAI announced deals with Mubadala, G42 and the Abu Dhabi government during the same visit, deepening a national footprint. DataStudios |
Compute Oracle is also embedding OpenAI's ChatGPT and Codex in its apps. At Oracle AI World the company paired the Google TPU news with native OpenAI features and Snowflake access on OCI. TheNextWeb |
Deals Cohere is in talks for up to $3 billion at a reported $20 billion valuation. The enterprise model vendor is raising a Series E, with investor interest described as strong but terms unconfirmed. PYMNTS |
Deals Hyphenate raised $45 million for finance-team agents. The Series A, led by Nexus and B Capital, backs agents aimed at the controller's office. New Economies |
Governance Gartner warns the AI layoff may reverse. Its projection that 30% of AI-displaced workers get rehired by 2029, at higher cost, is a caution for anyone cutting staff to a tool. ITPro |
Product OpenAI will retire Agent Builder and the Evals API on November 30. Teams that built on the preview tools must migrate, a reminder that vendor roadmaps move faster than enterprise ones. AIToolsRecap |
Policy Anthropic's updated usage policy takes effect November 12. The revised terms tighten rules around high-risk and agentic deployments, worth a read before you scale. AIToolsRecap |
Infrastructure Google is building its own TPU cloud with Blackstone. The venture, reported as Crux AI, would sell TPU capacity directly and complicates Oracle's claim to be first outside Google. The Stack |
|
| Quick HitsThe wider field, one line each. Raises | 01 | Pathos AI raised a $237M Series E led by Bristol Myers Squibb for AI drug development.Presspool | | 02 | AuxoAI raised about $21M in a Series A led by NTTVC for forward-deployed enterprise AI.Presspool | | 03 | Omnichat raised a $10M Series A led by Openspace for its AI agent platform.AI Funding |
|
Ships and deals | 04 | Oracle added Snowflake access and native ChatGPT and Codex features to OCI alongside the Google TPU news.TheNextWeb | | 05 | Dell said more than 6,500 enterprises now use its AI Factory with NVIDIA.AI Agents Directory | | 06 | Sema4.ai shipped Enterprise AI Agent Platform 3.0, adding agent sharing and cost controls.AI Agents Directory |
|
Rules and people | 07 | Apple reports fiscal Q4 after the close, the first call led by new CEO John Ternus.AppleInsider | | 08 | Senator Elizabeth Warren asked Anthropic to answer questions on its IPO, including its Amazon and Broadcom ties, by November 3.TheNextWeb | | 09 | eBay said it will cut about 9% of staff, and EA said it will shed 400 roles and close BioWare.InformationWeek | | 10 | Microsoft Ignite runs November 17 to 20, the next big enterprise AI roadmap stop.AIToolsRecap |
|
| From Spearhead · AI Bootcamp Get your leadership team AI-ready, in a day.The AI Bootcamp is a one-day, in-person session designed for your industry, your organization, and your use cases. Your leaders understand AI, build AI agents, envision outcomes, and start building a new work culture. |
| The Number $42.3B Worldwide spending this year on AI-optimized cloud infrastructure, nearly double 2025. Gartner forecasts 96% growth in AI-optimized infrastructure-as-a-service this year, from about $21.5 billion in 2025. Rented capacity bills by use, and the meter runs every day an agent is in production. More than half of that spend is now for running models rather than training them. Gartner | Counter-Signal Markets A slowdown that pays buyersMost of the coverage reads the quarter as bad news. For customers it may be the opposite. A vendor whose growth rate fell from 144% to 47% in a quarter has every reason to fight for the next contract, and the filing shows it already is: about $600 million of revenue went to lower prices and contract discounts. Capacity problems can also be bought down, and Anthropic reported $146.6 billion in liquidity at quarter end. Fortune The catch for buyers is the hook that comes with the discount. Anthropic reports net revenue retention near 900%, a figure the filing does not define and that also measures how fast existing accounts' bills grow. More than half of its $100,000-plus customers now use both its platform and its own apps. Take the lower price, and read the renewal terms twice. TechCrunch | From the Field Every enterprise AI contract is a bet on someone else's cost structure. This week one of those structures landed in an SEC filing. Most of the AI buyers we work with have never seen their model vendor's cost of goods. They negotiate on token price, seat count and a security questionnaire, and assume the supplier will hold the same terms next year. That was a reasonable bet while every lab ran on private money and a growth story. A prospectus changes it. The numbers are public now, and once it lists, the vendor answers for them every quarter. Model churn worries me more than the loss. When most of a vendor's revenue comes from models less than six months old, last spring's evaluation is already stale. The teams that handle this well keep a small regression suite for each workload and rerun it whenever the model underneath changes. The rest find out from their users. Whichever lab you buy from, write the contract as if a quarter like this one could happen again. Let's create value with AI, AK | Work with Spearhead AI Bootcamp One day to get leaders fluent enough to decide. |
|
| The Agentic Enterprise Know more about AI than 95% of your peers. Every morning. OpenAI, Anthropic, NVIDIA, AWS, Azure, Databricks and Snowflake are Spearhead technology partners (Microsoft and Amazon through Azure and AWS); Google, Oracle, Apple, Broadcom, Dell and ADNOC are not. Produced using Anthropic’s Claude with editorial direction from AK. © 2026 Spearhead. All rights reserved. |
|